// ITS FOSS — LINUX & OPEN SOURCE
Linus Torvalds Says He Likes Using AI, While His New Kernel Branch Credits Claude Code
Linus Torvalds has a new branch in his own kernel repository named noref-filemap-read-with-llm.
It holds an eight-patch series built on Linux 7.3-rc6 plus fixes, and its commit message gives no kernel release target for merging it.
While the commit message is long and technical, the change itself is easy to describe.
When many threads read the same file, the kernel serves them from the page cache, its in-memory copy of recently read data. Today, each read takes a hold of the cached data and releases it afterwards.
Every thread updates the same counter to do so, resulting in the threads tripping over each other. The issue dates to 2024, when Matthew Wilcox flagged it, citing a customer's workload.
Linus' series of patches skips that step. To make that safe, memory that leaves the cache is not handed out for reuse until a short period has passed, so a reader can copy data straight out without holding anything.
The cost is that more memory sits idle for a moment.
The payoff, measured on his own laptop with some security protections switched off, is that eight threads reading one file went from about 10 million reads per second to between 30 and 33 million.
Linus ran the measurements himself, but for the rest, the commit message credits Claude Code, working under his direction. That includes a draft of the message itself, the code and test programs, and the review behind its list of known problems.
This is not the first time he has used AI on kernel work. In August, he used it to help fix a bug in the Intel Xe graphics driver.